Episode 93 - Agentic CRM Webinar Full Recording

Welcome to Episode 93 of the Retention Blueprint. 

Last Thursday, instead of writing about retention, I taught it. Live.

467 of you registered. 156 joined on the day. Thank you, to everyone who came, asked questions, and stayed to the end.

The session was called Activating Agentic CRM 3.0: how to build one-to-one CRM you can actually govern. 

If you missed it, the full recording is now on YouTube. 

👉 https://www.youtube.com/watch?v=5DdNogjS7zc 

This week, I want to give you the short version. The five ideas I think matter most.

1. The gap is bigger than most leaders realise

Your data is individual. 

Your marketing is still built in segments.

You have behavioural signals, engagement patterns, lifecycle state, real-time attributes. And then you take all of that richness and send one message, lightly personalised, to a hundred thousand people.

Lifecycle programs work. Personalisation works. But their impact is capped, because every customer is different in every moment, and the average customer is only ever a fraction of your base.

CRM 3.0 closes that gap. You use individual signals to decide the next best action for every customer, one at a time, every time.

2. The principle that unlocks it: pointers, not payloads

This is the one idea to take away if you take nothing else.

The agent reads a customer's behavioural state. It picks an approved action. Then it emits a short, structured instruction: this customer, this action, this channel, this message, this creative, now.

It never stores content. It never writes content. It selects.

That single design choice is what makes everything else, governance included, possible.

3. It isn't unauditable. You're auditing the wrong layer

The number one objection I hear is "if there's no journey to look at, how do we audit it?"

In CRM 1.0, your audit trail was the send list. In CRM 2.0, it was the journey canvas. In CRM 3.0, there is no single canvas, because every customer gets a different path in real time.

So leaders conclude it can't be governed. They're wrong. They're just auditing the wrong layer.

You don't audit a million journeys. You audit the envelope (the rules the agent can't break), the ledger (every decision it made, reconstructable) and the distribution (what the whole base actually received). Boring, defensible, and exactly what your regulator wants to see.

4. Netflix and Uber already do this

This is not theory. Both have published how they engineer their messaging, and it maps straight onto CRM 3.0.

Netflix runs a "slow" planner that builds a weekly pacing plan per user, and a "fast" real-time engine that decides what to send in the moment. The decision is separated from the send. Frequency is governed. 

Uber uses contextual bandits and a real-time feature store, so a change in your context updates the offer instantly.

Neither uses a visual journey builder. They build big menus of approved content and let agents read real-time signals and choose per person. That is CRM 3.0, at scale.

5. The blocker is never the AI. It's the data (but you don't need perfect data)

These projects don't fail on the model. They fail on the data underneath.

But you don't need a perfect 500-event schema to begin. For one moment of truth, you often need just 10 to 15 events, sometimes from a single source. Pick one moment. Get those events right. Prove it on a slice of your base. Then expand.

The teams winning at this started narrow and earned the right to scale.

Final Thoughts

Gartner expects 60% of brands to use agentic AI for one-to-one interactions by 2028. The shift is already happening, and the gap between the teams who start and the teams who wait is widening now.

The good news: most enterprises already own most of the architecture. The work is in defining the action space, building the governance, and creating the content. None of it is out of reach.

If you want the full picture, the architecture, the analogies and the governance model, watch the recording here

👉 https://www.youtube.com/watch?v=5DdNogjS7zc 

Until next time,

Tom

Want help building this? Book a call.